Custom template for CAST
logo



Start with a short introduction about what is going on in this document

Introduction


Introduction to PDCA, CAST and the methods contained within this document go here.

Background


Explain some background info that is needed. Change header colours to purple?

Arps Decline models used to forecast historically and which models will be analyzed here

Modified Hyperbolic

Explanation of that equation and a definition using code..?

Sample Production Data

Step through explanation of 2 stream sample data included. Plot both oil and gas production. Pick one well to

Plots

PDCA

Explain the background of what it is and how it can be helpful in forecasting, especially to handle uncertainty. Define PDCA here! Explain briefly bayesian methods of achieving this? Priors and posteriors

Monte Carlo Method

Explain the place of the monte carlo method as our first try at estimating the posterior ditribution, showing parameter distributions as defining our prior and explaining how they are each sampled

MCMC Method

Explain here and have links to jump around document? Or explain when we get to it?

CAST

Forecast Synthesis

Explain process of forecast sythesis here, and show how it is carried out for the modified hyperbolic. Show arps substitution, and explain

Loss Functions

How do we compare the synthesis forecasts against our actuals and determine what is a good fit and what is a poor fit? S

Soft L1 Loss

Define and apply soft L1 loss and show plot of a forecast against actuals colour by individual cost

PDCA Loss

Explain this loss function, where it’s from, and show a comparison of application to the Soft L1 loss

Cost Evaluation

Show evaluation of the loss function to come up with average cost parameters for each forecast Show plots coloured by avg cost

Best Fit

Show best fit using either loss function and explain how this may be a good fit for wells with good production history, we still want to create a probabilistic outcome for each well to capture forecast uncertainty

Show top 10% of functions, etc. using either loss function, and show that the Monte Carlo method returns adequate forecasts for good prod history, but not short history wells.. What’s next?? MCMC

Markov Chain Monte Carlo (MCMC)

Explain concept or link above if put there.

Burn-In

Show MC method as burn-in to determine best fit forecast to apply PDCA loss function to

Jumping function and distributions

Results

Probabilistic forecasts

Creation of P10, P50, Avg, and P90 forecasts

 




Darcy Redpath
Petronas Canada